CadArena: Natural Language → CAD Drawings
Capstone Graduation Project Awarded A+ (99%) at Beni-Suef National University
Beni-Suef National University
Beni-Suef National University Capstone 2026 · Capstone Project 2026
Cognitive CAD platform translating natural language engineering commands into precise vector CAD drawings (DXF format) using LLMs, RAG, and automated ezdxf Python engines.
Cognitive CAD Automation Platform
Executive Overview
CadArena is an end-to-end Generative AI platform designed to transform natural language engineering instructions into precise vector CAD drawings (DXF format) using Large Language Models. Engineers describe mechanical or architectural components in plain text, and CadArena generates accurate, production-ready vector drawings in real time.
Developed as my Capstone Graduation Project at the Faculty of Computers & Artificial Intelligence, Beni-Suef National University, CadArena was awarded an A+ grade (99%). The platform bridges cutting-edge LLM reasoning with classic computer-aided design constraints to streamline engineering design workflows.
Project Presentation Deck (54 Slides)
Full original presentation deck defended at Beni-Suef National University. Click thumbnails or use controls to browse.
Slide Strip Preview (Click to view)::
Architecture & Workflow
Engineering infrastructure from text prompt to vector DXF file
Natural Language Input
Engineers input plain text prompt requests (e.g. "Draw a 200x100mm flange with 4 M8 holes"). Input is sanitized and tokenized.
LLM Intent Parsing
Custom-prompted LLMs (Ollama / OpenAI) parse text into structured geometric primitives (lines, circles, arcs, layers, constraints).
RAG CAD Constraints
RAG Knowledge Layer indexes standard CAD conventions, drawing tolerances, and DXF layer rules to enforce valid engineering geometry.
ezdxf Vector Compiler
Automated Python DXF compiler converts parsed geometric entities directly into valid ezdxf vector drawing structures.
Real-Time WebSockets Stream
WebSockets stream vector drawing progress continuously, updating the interactive React canvas in real time.
AI CAD Iteration
Users modify geometry via follow-up natural language instructions while preserving existing drawing layers and dimensions.
CadArena Demo Video
Watch CadArena translate natural language prompts into precise vector CAD drawings in real time.
Tech Stack Breakdown
Core tools and frameworks powering CadArena
Python
Core Backend & Data Generation
FastAPI
High-performance Async REST API
React
Interactive Canvas UI & State
Ollama
Local LLM Inference Engine
LLMs
Command & Intent Parsing
RAG
CAD Standards Knowledge Layer
WebSockets
Bi-directional Real-Time Streaming
PostgreSQL
Drawing Metadata & History
Docker
Containerized Deployment
Awarded A+ Grade (99%) Capstone Honor
Evaluated at Beni-Suef National University — Faculty of Computers and Artificial Intelligence.